EpiGraph: A Knowledge Graph and Benchmark for Evidence-Intensive Reasoning in Epilepsy
Researchers have introduced EpiGraph, a large-scale knowledge graph and benchmark designed to enhance evidence-intensive clinical reasoning in epilepsy diagnosis and treatment. This initiative addresses the complexity of integrating heterogeneous clinical data, including biosignal patterns, genetic mechanisms, pharmacogenomics, and patient outcomes. EpiGraph synthesizes information from 48,166 peer-reviewed papers and seven clinical resources, creating a heterogeneous graph with over 24,000 entities and 32,000 evidence-grounded triplets across five clinical layers. Accompanying this graph is EpiBench, a benchmark framework defining five clinically motivated tasks such as clinical decision-making, EEG report generation, and treatment recommendation. The study evaluated six Large Language Models (LLMs) under standard and Graph-RAG settings, revealing that integrating EpiGraph consistently improved performance across all tasks. Notably, pharmacogenomic reasoning saw significant gains of 30-41%. These findings highlight the potential of structured medical knowledge to augment LLM capabilities in real-world neurological settings, offering a robust tool for advancing precision medicine and clinical AI applications.
Wire timeline
EpiGraph: A Knowledge Graph and Benchmark for Evidence-Intensive Reasoning in Epilepsy
Researchers have introduced EpiGraph, a large-scale knowledge graph and benchmark designed to enhance evidence-intensive clinical reasoning in epilepsy diagnosis and treatment. This initiative addresses the complexity of integrating heterogeneous clinical data, including biosignal patterns, genetic mechanisms, pharmacogenomics, and patient outcomes. EpiGraph synthesizes information from 48,166 peer-reviewed papers and seven clinical resources, creating a heterogeneous graph with over 24,000 entities and 32,000 evidence-grounded triplets across five clinical layers. Accompanying this graph is EpiBench, a benchmark framework defining five clinically motivated tasks such as clinical decision-making, EEG report generation, and treatment recommendation. The study evaluated six Large Language Models (LLMs) under standard and Graph-RAG settings, revealing that integrating EpiGraph consistently improved performance across all tasks. Notably, pharmacogenomic reasoning saw significant gains of 30-41%. These findings highlight the potential of structured medical knowledge to augment LLM capabilities in real-world neurological settings, offering a robust tool for advancing precision medicine and clinical AI applications.
cs.AI updates on arXiv.org